An experimental analysis of feature selection and similarity assessment for textual summarization

Abstract

Since the access to information is increasing every day, and we can quickly acquire knowledge from many sources such as news websites, blogs, and social networks, the capacity of processing all this information becomes increasingly difficult. So, tools are needed to automatically extract the most relevant sentences, aiming to reduce the volume of text into a shorter version. One alternative to achieve this process while preserving the core information content by using a process called Automatic Text Summarization. One relevant issue in this context is the presence of typos, synonyms, and other orthographic variations since some extractive techniques are not prepared to handle them. This work presents an evaluation of different similarity approaches to minimize these problems, selecting the most appropriate sentences to represent a document in an automatically generated summary.